基于改进AKAZE的特征匹配算法及应用

Feature Matching Algorithm of Huizhou Architecture Based on Improved AKAZE

徽派建筑群的代表性构件样本量大,且由于相同类型构件外观造型相似,仅花纹图案有所区别,所以导致难以判断所采集图像的归属。提出了G-AKAZE方法用于特征匹配,有效提升了匹配速度和精确度,并将其用于徽派建筑图像数据的匹配。首先非线性方法构造尺度空间,并用快速显示扩散数学框架FED来快速求解偏微分方程,再用Hessian矩阵进行特征点检测,根据特征点获取主方向并旋转图像,通过采样网格的像素完成尺度自适应,最后将图像网格化,去除误判点完成特征匹配。此方法能快速且准确地对目标图像进行特征匹配,实验使用前期采集的徽派建筑图像数据,在匹配速度和匹配对数两方面的表现优于同类特征匹配方法。

The representative components of the Huizhou architectural complex have a large sample size and it is difficult to determine the belonging classification of the collected images.In this paper,g-AKAZE method is proposed for feature matching,which can effectively improve the matching speed and accuracy,and it is used for the matching of Huizhou architecture image data.First,a non-linear method is used to construct the scale space,and the fast differential display mathematical framework FED is used to quickly solve partial differential equations.Then Hessian matrix is used to detect feature points.The main direction is obtained based on the feature points and then rotate the image.The pixels of the grid are sampled to complete the scale adaptation.Finally,the image is meshed to remove the misjudgment points to complete the feature matching.This method is used to match the features of the target images quickly and accurately.The experiment uses the image data of Huizhou Architecture collected earlier,and the matching speed and matching logarithm are better than those of the similar method.